A network of genetic repression and derepression specifies projection fates in the developing neocortex

Abstract

Neurons within each layer in the mammalian cortex have stereotypic projections. Four genes—Fezf2, Ctip2, Tbr1, and Satb2—regulate these projection identities. These genes also interact with each other, and it is unclear how these interactions shape the final projection identity. Here we show, by generating double mutants of Fezf2, Ctip2, and Satb2, that cortical neurons deploy a complex genetic switch that uses mutual repression to produce subcortical or callosal projections. We discovered that Tbr1, EphA4, and Unc5H3 are critical downstream targets of Satb2 in callosal fate specification. This represents a unique role for Tbr1, implicated previously in specifying corticothalamic projections. We further show that Tbr1 expression is dually regulated by Satb2 and Ctip2 in layers 2–5. Finally, we show that Satb2 and Fezf2 regulate two disease-related genes, Auts2 (Autistic Susceptibility Gene2) and Bhlhb5 (mutated in Hereditary Spastic Paraplegia), providing a molecular handle to investigate circuit disorders in neurodevelopmental diseases.

Cortical neurons project to specific targets located locally within the cortex or in distant subcortical regions, depending on the neuron’s birthdate and laminar position. Neurons in layer 5 project subcortically to targets that include the spinal cord, whereas neurons in layers 2/3 typically project corticocortically, including across the corpus callosum (CC) (1–3). Four key genes—Fezf2, Ctip2, Tbr1, and Satb2—are involved in specifying projection neuron identity (3–7). Fezf2 and Ctip2 regulate the identities of layer 5 subcerebral projection neurons (SCPNs), whereas Satb2 regulates the fate of callosally projecting neurons (Fig. 1O). Without Fezf2, SCPNs up-regulate Satb2, switch fate, and extend axons across the CC (8), suggesting that Fezf2 represses Satb2 and the acquisition of a callosal identity. Conversely, without Satb2, callosal neurons up-regulate Ctip2 (but not Fezf2) and project subcortically (9, 10). Biochemical experiments revealed that Satb2 binds to and represses the Ctip2 locus (9) but have not ascertained whether the fate switch in Satb2 mutant neurons is a direct consequence of Ctip2 upregulation.

Fig. 1.

Loss of Ctip2 partially rescues callosal projections in Satb2 mutant neurons. (A–C) LacZ staining of P0 whole-mount brains (ventral view) reveals aberrant β-gal⁺ axons (blue) in the pyramidal tract (arrow) in Satb2^LacZ/flox;Emx1-Cre mutants (B) but not in control mice (A) or Ctip2^−/−;Satb2^LacZ/flox;Emx1-Cre double knockouts (C). (D) The majority of β-gal⁺ callosal axons in Satb2 conditional mutants fail to project across the CC at P0. (E and F) Instead, these axons project subcerebrally into the cerebral peduncle (cp) and thalamus (Th). (G) A small number of β-gal⁺ axons are seen in the pyramidal tract (PyT). (H) Coronal sections of Ctip2^−/−;Satb2^LacZ/+ mice reveal β-gal⁺ axons crossing normally at the CC. (I and N) β-gal⁺ callosal axons in Satb2;Ctip2 double mutants are seen at the CC (arrow) at P0. Defasciculated β-gal⁺ axons also descend subcortically into the striatum (J), along the cerebral peduncle (K and L) but not to the level of the pyramidal tract (M). (O) Model outlining genetic interactions between Fezf2, Ctip2, Satb2, and Tbr1 described previously during cortical projection neuron fate specification. Question marks denote interactions that are not clearly understood. (Scale bar in D, 200 μm.)

Layer 6 corticothalamic neurons and layer 5 SCPNs show a similar mutual repression between Tbr1 (layer 6) and Fezf2 (layer 5) (3–7). Deep layer neurons in Tbr1^−/− mice up-regulate Fezf2 and project inappropriately to the pons, while Fezf2^−/− neurons up-regulate Tbr1 and form thalamic projections (Fig. 1O). Thus, Fezf2 represses both Tbr1 and Satb2, which repress corticothalamic and callosal fates in SCPNs. The nature of these repressive interactions, however, is unknown. Here we analyze mice bearing mutations in Fezf2, Ctip2, and Satb2 to examine the genetic relationships between these pathways. Using in vivo electroporation to rescue callosal axons in Satb2 mutants, we identified downstream targets of Satb2 that play key roles in the formation of callosal projections.

In this study, data obtained from double knockouts of Satb2, Fezf2, and Ctip2 (together with previously published data) provide evidence for a genetic model of cortical projection neuron fate determination during early corticogenesis (Fig. S1A). In Fezf2-expressing neurons, expression of Tbr1 and Satb2 is repressed, which results in a suppression of corticothalamic and callosal fates, respectively, and axon extension along the corticospinal tract (CST). Fezf2 mutant neurons fail to repress Satb2 and Tbr1, thus their axons cross the CC and/or innervate the thalamus inappropriately.

In callosal projection neurons, Satb2 represses the expression of Ctip2 and Bhlhb5, leading to a repression of subcerebral fates. Interestingly, we found that Satb2 promotes Tbr1 expression in upper layer callosal neurons, and Tbr1 expression in these neurons is required for callosal specification (not corticothalamic specification). Our data suggest that the projection fates of deep layer and upper layer cortical pyramidal neurons are specified by complex and mutually inhibitory genetic interactions between postmitotic determinants during late embryonic development.

Results

Visualization of Callosal and Subcortical Axonal Connections in Satb2 and Fezf2 Mutants.

We visualized callosal axons using mice in which LacZ was targeted to the Satb2 locus (9). In embryonic day (E) 18.5 control Satb2^LacZ/+ mice, β-galactosidase positive (β-gal⁺) axons cross the CC and form corticocortical connections (Fig. S2 B and O) but do not extend subcortically (Fig. S2 A and O). As shown previously, β-gal⁺ axons in E18.5 Satb2^LacZ/LacZ mutants fail to cross the CC and instead descend subcortically through the striatum and along the cerebral peduncle and to the thalamus (Fig. S2 C, D, and P). Thus, without Satb2, callosal neurons seem to be respecified as subcortical projection neurons (9, 10). Satb2^LacZ/LacZ mice die within a few hours after birth, which precludes the analysis of axonal connectivity at postnatal stages. Here we also use mice bearing a conditional allele of Satb2 (Satb2^flox/flox, generated in the laboratory of R.G.) bred with mice that express Cre recombinase under the Emx1 promoter (Emx1-Cre) to produce viable, cortex-specific Satb2 knockouts. Mice carrying Emx1-Cre also carried the Satb2^LacZ allele, enabling us to generate conditional Satb2^LacZ/flox;Emx1-Cre knockout mice in which the axons of neurons that normally express Satb2 neurons are marked by β-gal (Fig. 1B, D–G). A detailed description of Satb2 conditional knockout mice is in preparation.

The insertion of a placental alkaline phosphatase (PLAP) cassette into the Fezf2 locus (11) enabled us to visualize the projections of deep layer cells that normally express Fezf2. PLAP⁺ axons in Fezf2^PLAP/+ controls are observed in the pyramidal tract (Fig. 2 A and E) and thalamus (Fig. S4I), whereas labeled axons are absent from the CC, as previously described (8, 11). At P0, PLAP⁺ axons descend subcerebrally and branch toward the thalamus (Fig. S3B) and cerebral peduncle (Fig. S3B) but do not extend past the pons. At postnatal day (P) 4, Fezf2^PLAP/PLAP mutants show a striking reduction in the number of PLAP⁺ axons that extend into the pyramidal tract at the level of the hindbrain (Fig. 2 B and F). In addition, a subset of PLAP⁺ axons cross the CC (Fig. 2 I and J), consistent with the interpretation (Fig. S1B) that at least some neurons that normally express Fezf2 acquire a callosal identity in the absence of Fezf2, possibly owing to acquisition of Satb2 expression (8).

Fig. 2.

Loss of Satb2 does not rescue PLAP⁺ axons in Fezf2^PLAP/PLAP mutants. (A–D) Whole-mount PLAP staining of Fezf2^PLAP/+ heterozygous mice (A) and Satb2 conditional mutants (C) reveals that PLAP⁺ axons extend into the pyramidal tract (PyT). (B) Fezf2^PLAP/PLAP mutants show a dramatic decrease in PLAP staining in the pyramidal tract. (D) In Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants, there is a complete loss of PLAP staining in the pyramidal tract. (E–H) PLAP staining of sagittal sections from the brains shown in A–D. PLAP⁺ axons in heterozygous Fezf2^PLAP/+ controls (E) and Satb2^LacZ/flox;Emx1-Cre mutants (G) project subcortically through the pyramidal tract (PyT) and pyramidal decussation (pd). (H) PLAP⁺ axons in Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants fail to enter the pyramidal tract. (Scale bar, 200 μm.) (I and J) In Fezf2 mutants, PLAP⁺ axons are present in the CC in both sagittal (I) and coronal (J) sections. (K) Both PLAP⁺ (red) and β-gal⁺ (green) axons fail to cross the CC in Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants. (Scale bars, 200 μm.)

Interactions Between Fezf2 and Satb2 in the Regulation of Subcortical Identity.

We tested the role of Satb2 in the respecification of deep layer fates by generating Fezf2;Satb2 double mutants, predicting that PLAP⁺ axons should fail to extend across the CC. Indeed, analysis of Fezf2^PLAP/PLAP;Satb2^LacZ/LacZ mice revealed a loss of PLAP⁺ axons in the CC at P0 (Fig. 2K), providing strong evidence that the acquisition of a callosal fate of Fezf2 mutant neurons is due to up-regulation of Satb2. Visualization of PLAP⁺ axons in the sagittal plane showed axons projecting subcortically through the striatum and exiting through two branches, one directed toward the thalamus (Fig. S3 B and C) and the other into the cerebral peduncle (Fig. S3 B and C), although the latter axons fail to reach the hindbrain (Fig. S3B, asterisk). The β-gal⁺ axons of Satb2-expressing neurons in the same section follow a similar tract, although they extend past the peduncle toward the pons (Fig. S3 A and C) but fail to reach it.

This difference in the extent of subcortical innervation by PLAP⁺ vs. β-gal⁺ axons suggested that loss of Satb2 might affect the formation of subcerebral projections by layer 5 neurons, which normally express Fezf2 but not Satb2. To ascertain whether mice lacking both Fezf2 and Satb2 simply show a developmental delay in the growth of subcerebral projections, we analyzed PLAP staining at P4 in Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants. Whole mounts revealed a complete loss of PLAP⁺ axons in the pyramidal tract (Fig. 2D), whereas littermates lacking only Fezf2 retain some PLAP staining of pyramidal tract axons (Fig. 2B). These data suggest the surprising possibility that Satb2 plays a role in the normal targeting of SCPNs.

If this hypothesis is correct, subcerebral axons in Satb2 mutants should display defects in their projection patterns. We therefore examined the pattern of PLAP labeling in mice lacking Satb2 and carrying one copy of the Fezf2^PLAP allele. In Fezf2^PLAP/+;Satb2^LacZ/LacZ mutants at E18, a thin tract of PLAP⁺ axons extends subcortically toward the cerebral peduncle but does not reach the pyramidal tract (Fig. S4F). In conditional Satb2 mutants (Fezf2^PLAP/+;Satb2^flox/flox;Emx1-Cre) at P4, PLAP⁺ axons reach past the pons to the pyramidal decussation (Fig. 2 C and G), but the axons fail to enter the spinal cord (Fig. S4 G and H). To determine whether this reflects a failure of corticospinal innervation or a developmental delay, we immunostained conditional Satb2 knockouts at P21 (when PLAP can no longer be detected in our animals) using antibodies against protein kinase C γ (PKCγ), which labels the CST (12). These studies revealed a complete loss of PKCγ staining in the spinal cord, consistent with a loss of corticospinal connections.

These analyses of PLAP⁺ axons in Satb2 mutants yield the unexpected conclusion that Satb2 is required for the normal growth of subcerebral axons to their targets in the spinal cord. How might Satb2 exert this effect? One possibility is that Satb2 plays a non-cell autonomous role in subcerebral neurons development. However, our prior studies have shown that Satb2 and Ctip2 are normally coexpressed in layer 5 neurons during a brief period early in corticogenesis (E13.5–E14) (9). Thus, although Satb2 may be expressed only transiently in subcerebral neurons, this window of expression may play a critical role in regulating the expression of key genes such as axon guidance molecules. For example, Satb2 mutants show reduced expression of Unc5H3, a well-characterized axonal guidance receptor (9). In Unc5H3 mutants, subcerebral axons fail to descend past the decussation into the spinal cord (13), raising the possibility that Satb2 knockouts display a similar phenotype due to reduced expression of Unc5H3. An alternative possibility is that Satb2 may play a non-cell autonomous role in guiding subcerebral axons to the spinal cord. For example, Satb2 is expressed in interneurons and in the spinal cord, and it is possible that this expression is required for corticospinal axon guidance. However, in our analyses of conditional Satb2 mutants (in which expression in interneurons is maintained), we observe the same phenotype as the straight Satb2 knockouts. Therefore, it is unlikely that Satb2 expression in interneurons is essential for guiding subcerebral axons to the spinal cord.

Genetic Interactions Between Satb2 and Ctip2 in the Formation of Subcerebral Projections.

Previous studies using Ctip2 knockout mice suggest that Ctip2 is required for the normal formation of subcerebral projections, which are defasciculated and terminate prematurely in these mutants (1). Ctip2^−/− mice do not carry a reporter allele (1), therefore analysis of axonal projections in these animals has been limited to using dye tracers. Because Fezf2 expression is not altered in Ctip2^−/− mutants (Fig. S4A) and because Fezf2 and Ctip2 seem to be coexpressed in deep layer neurons (8, 14), we used Fezf2^PLAP as a marker for subcerebral projections. As expected, PLAP⁺ axons in Ctip2^+/−;Fezf2^PLAP/+ heterozygous brains extend subcortically through the striatum; some extend into the thalamus, whereas others are present in the cerebral peduncle (Fig. S4I) and pyramidal tract (Fig. S4J). In Ctip2^−/− mutants carrying a Fezf2^PLAP allele, defasciculated PLAP⁺ axons pass through the striatum and extend into the thalamus or cerebral peduncle (Fig. S4K) but do not reach the pons (Fig. S4K, Inset). This pattern is similar to that seen by dye-labeling subcerebral axons in Ctip2 mutants (1).

Our model of the genetic interactions that confer projection neuron fates predicts that in the absence of both Ctip2 and Satb2 (Fig. S1F), PLAP⁺ SCPNs should form projections similar to those seen in Ctip2 single mutants (Fig. S1C). We used PLAP staining to trace the projections of these neurons in Ctip2;Satb2 double mutants (Ctip2^−/−;Fezf2^PLAP/+;Satb2^LacZ/flox;Emx1-Cre) (Fig. S3 D–F). PLAP⁺ axons (in red) are absent from the CC and can be seen subcortically at the level of cerebral peduncle (Fig. S3E). PLAP⁺ axons are absent from the thalamus in sagittal sections (Fig. S3D, asterisk), consistent with the prediction (Fig. S1F) that layer 5 neurons that lose both Ctip2 and Satb2 expression will express only low levels of Tbr1. These neurons, however, continue to express Fezf2 and Bhlhb5, both of which impart a subcortical identity to the neurons, as evidenced by PLAP⁺ axons in the cerebral peduncle. However, without Ctip2, these neurons are unable project past the pons.

Loss of Satb2 Rescues Ctip2 Expression in Fezf2^PLAP/PLAP Neurons.

Ctip2 expression is down-regulated in layer 5 of Fezf2 mutants (11) (Fig. 3B), suggesting that Ctip2 may be a downstream target of Fezf2. However, because Fezf2 normally represses Satb2 genetically (8), and Satb2, in turn, represses Ctip2 (9) (Fig. 3C), it is possible that the loss of Ctip2 expression in Fezf2^PLAP/PLAP mutant neurons is an indirect consequence of the up-regulation of Satb2, rather than the result of direct transcriptional control of Ctip2 by Fezf2. If the regulation is indirect via Satb2, we predict that Ctip2 expression will be restored in layer 5 neurons in Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants (Fig. S1E). Indeed, we do observe strong Ctip2 expression in layer 5 neurons in double mutants (Fig. 3D), suggesting that Fezf2 regulates Ctip2 indirectly by repressing Satb2.

Fig. 3.

Ctip2 expression is restored in layer 5 neurons of Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants. (A) In WT mice, the expression of Ctip2 protein is highest in layer 5 neurons, moderate in layer 6, and largely absent from the upper layers (2–4). (B) Fezf2 mutants show a severe reduction of Ctip2 expression in layer 5, whereas expression in layer 6 is increased. (C) Satb2 mutant mice show increased Ctip2 expression in layer 5, 6, and the upper layers. (D) Ctip2 expression is rescued in layer 5 of Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants. (A–D) Each panel is a composite of tiled confocal images to form the full figure. (Scale bar, 50 μm.) (E–G) Graphs depict the percentages of Ctip2⁺ neurons in the upper layers (E), layer 5 (F), and layer 6 (G) in various genotypes. Error bars are average ± SEM (n = 3 different animals). Confidence levels were calculated using a t test (*P < 0.5; **P < 0.05; ***P < 0.005, relative to controls).

We counted the number of Ctip2-expressing cells across the cortical layers, first in single knockouts for Fezf2 or Satb2. In layer 6 (Fig. 3G), the number of Ctip2⁺ cells increased from 29% ± 1.5% in control mice to 39% ± 1.8% (P < 0.006) in Fezf2 mutants and 47% ± 5.4% (P < 0.01) in Satb2 conditional mutants. In layer 5, the total number of Ctip2⁺ cells decreased in Fezf2 mutants (57% ± 7% in controls vs. 10% ± 3% in mutants, P < 0.001) but increased to 83% ± 4% in Satb2 mutants (P < 0.01) (Fig. 3F). This suggests that the loss of Ctip2 in layer 5 of Fezf2 mutants might be due to the acquisition of Satb2 expression (Fig. S4C) (8). Satb2 also represses Ctip2 expression in the upper layers, as evidenced by the dramatic increase in Ctip2⁺ cells from 11% ± 1.7% in controls to 77% ± 9.5% (P < 0.05) in Satb2 mutants (Fig. 3E).

To ascertain whether Ctip2 is regulated directly by Fezf2 or indirectly by Satb2, we counted the numbers of Ctip2⁺ cells in Satb2^LacZ/flox;Fezf2^PLAP/PLAP;Emx1-Cre double mutants. The loss of Ctip2 expression in layer 5 neurons of Fezf2^PLAP/PLAP mutants (10% ± 3%) is rescued in double mutants (65% ± 1.8%; P < 0.0005) (Fig. 3F), providing strong evidence that at this stage in development, Fezf2 regulates Ctip2 indirectly by repressing Satb2.

Satb2 Regulates Bhlhb5 Expression.

Knowing that Fezf2 regulates Ctip2 indirectly by repressing Satb2, we asked whether other genes important for SCPN identity, such as Bhlhb5, are regulated similarly. Bhlhb5 is expressed by Ctip2⁺ neurons in layer 5 as well as in neurons in layers 2/3 (12). Bhlhb5 functions as an area-specific transcription factor that regulates the postmitotic acquisition of area-specific identities. In caudal motor cortex, Bhlhb5 null mice exhibit anomalous differentiation of corticospinal motor neurons, accompanied by failure of CST formation (12). The axons stop at the base of the pons and fail to enter the pyramidal tract (similar to Ctip2^−/− mutant axons at P0).

To ascertain whether Fezf2 regulates the expression of Bhlhb5, we immunostained Fezf2^PLAP/PLAP brains for Bhlhb5 protein. The number of Bhlhb5⁺ cells is significantly reduced in layer 5 of Fezf2 mutants (Fig. 4C) relative to controls (Fig. 4A) (64% ± 3% in controls to 12% ± 4.5% in mutants; P < 0.005). If this is due to the expansion of Satb2 expression in Fezf2 mutants (Fig. S4C), Bhlhb5 expression should be rescued in layer 5 neurons that lack Satb2. Indeed, the number of Bhlhb5⁺ cells in layer 5 in P4 mice is restored to 49% ± 3.9% in double mutants (Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre) (Fig. 4 D and H; P < 0.001 compared with Fezf2 single mutants). These data suggest that Fezf2 regulates Bhlhb5 expression in layer 5 indirectly by repressing Satb2, and that Satb2 functions as a repressor of Bhlhb5. Consistent with the latter interpretation, upper layer neurons of Satb2 mutant mice exhibited a robust increase in both the number of Bhlhb5⁺ neurons in layers 2–4 (Fig. 4G; control: 28% ± 7.8%; Satb2^LacZ/flox;Emx1-Cre conditional mutants, 63% ± 2.4%, P < 0.005) (Fig. 4B). Because Bhlhb5 is normally expressed by upper layer neurons, this regulation by Satb2 is unlikely to represent an on/off switch, but rather a mechanism that regulates the timing and levels of Bhlhb5 expression. Indeed, Bhlhb5 expression seems unchanged in the upper layers of Satb2 mutants at E18; the change is evident only by P4. This suggests that Satb2 likely regulates the timing of down-regulation of Bhlhb5 in upper layer neurons rather than an on/off switch for Bhlhb5 expression itself. It is also possible that Bhlhb5 is regulated by other factors expressed by upper layer neurons and that the final expression pattern reflects the action of multiple regulatory pathways.

Fig. 4.

Satb2 regulates Bhlhb5 expression in a layer-specific manner. (A) In controls, Bhlhb5 protein is expressed at high levels in layer 5 neurons and at low to moderate levels in the upper layers. (B) In Satb2^LacZ/flox;Emx1-Cre mutants, Bhlhb5 expression is significantly reduced in layer 5. (C) Expression in both layers 5 and 6 is decreased in the Fezf2 mutant cortex. (D) Bhlhb5 expression in layer 5 is rescued in Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants. (E and F) In Ctip2^−/− mutants (E) and Ctip2^−/−;Satb2^LacZ/LacZ double mutants (F), expression is increased in both layer 5 and the upper layers. (A–F) Each panel is a composite of tiled confocal images to form the full figure. (Scale bar, 50 μm.) (G–I) Graphs depict the percentages of Bhlhb5⁺ neurons in different genotypes in different cortical layers. Error bars depict average ± SEM (n = 3 different animals). Confidence levels were calculated using a t test (*P < 0.5; **P < 0.05; ***P < 0.005, relative to controls). (J–P) Ectopic overexpression of Bhlhb5 in upper layer neurons of WT animals can redirect their axons subcerebrally. (J–L) Sagittal sections of P4 control brains electroporated with GFP at E15 (when upper layer neurons are born) show (J) GFP⁺ neurons in the upper layers; (K) no GFP⁺ axons in the ipsilateral striatum (Str); and (L) GFP⁺ axons in the CC. (M–P) Sagittal sections of P4 brains electroporated with Bhlhb5 and GFP at E15 show (M) GFP⁺ neurons in the upper layers; and (N) robust GFP⁺ tracts extending through the striatum (Str) and (P) toward the cerebral peduncle (cp, arrows). (O) GFP⁺ axons are also observed in the CC. (Scale bar, 50 μm.)

Finally, we found that normal Bhlhb5 expression in some deep layer neurons also requires Ctip2. The percentage of Bhlhb5⁺ neurons in layer 6 is increased significantly in Ctip2^−/− mice (Fig. 4 E and I; control: 9% ± 4%, mutants: 70% ± 4.5%; P < 0.0005). However, in contrast to the severe reduction of Bhlhb5⁺ neurons in layer 5 of Fezf2 mutants, there is no significant change in the percentage of layer 5 neurons that express Bhlhb5 in Ctip2 mutants. Because Satb2 expression seems to be unaltered in the absence of Ctip2, these data are consistent with the interpretation that Satb2 is a primary regulator of Bhlhb5 in layer 5.

To determine whether high levels of Bhlhb5 expression are sufficient to direct axons toward subcortical targets, we coelectroporated expression constructs encoding Bhlhb5 and GFP into WT embyros at E15.5, a time when upper layer neurons are being generated. As expected, GFP labeled neurons are present in the upper layers in control (GFP only) (Fig. 4J) and Bhlhb5 plus GFP (Fig. 4M) electroporated brains. At P4, in control brains, GFP-labeled axons were confined to the ipsilateral and contralateral cerebral cortices (Fig. 4J), and no GFP⁺ axons can be observed subcerebrally in the striatum (Fig. 4K) or at the level of the cerebral peduncle. In these brains, GFP⁺ axons can be seen at the CC (Fig. 4L), indicating that as expected, electroporated neurons from upper layers may be sending the axons across the midline to the contralateral hemisphere. In contrast, the GFP⁺ axons of Bhlhb5-electroporated neurons extend beyond the striatal boundary into the internal capsule (Fig. 4N) and toward the cerebral peduncle (Fig. 4P, arrow). We do observe some GFP⁺ axons in the CC (Fig. 4O), which is probably because not all GFP⁺ cells express bhlhb5. These results suggest that Bhlhb5 by itself has a limited but real potential to direct axons to subcortical targets, less robust than that of Ctip2 (8) and similar to that observed after Sox5 electroporation at E15.5 (15, 16).

Fezf2, Ctip2, and Satb2 Regulate Tbr1 Expression in a Layer-Specific Manner.

Fezf2 and Tbr1 play important antagonistic roles in the regulation of layer 6 axons to the thalamus. Tbr1 mutants show a loss of corticothalamic projections, whereas Fezf2 mutants show a striking increase in thalamic innervation, most likely due to increased expression of Tbr1 (5, 7). In controls (Fig. 5A), the highest level of Tbr1 expression is observed in layer 6, with lower levels in layer 5 and scattered cells labeled in the upper layers. Consistent with previously published reports (5, 7), we found that the number of Tbr1⁺ neurons increases significantly in layer 5 and upper layer neurons of Fezf2 mutants (Fig. 5 E, G, and H) (control layer 5: 22% ± 4.7%, mutant layer 5: 54% ± 8.9% P < 0.05; control upper layers: 23% ± 7%, mutant upper layers: 43% ± 3%; P < 0.05) but is not significantly changed in layer 6 (Fig. 5I). Fezf2 mutants also show a robust band of PLAP fibers entering (Fig. 6F) and innervating the anterior thalamus (Fig. 6M), which might be a consequence of increased Tbr1 expression in layer 5. Interestingly, Ctip2^−/− mutants also show a significant increase in the number of Tbr1⁺ layer 5 neurons (most obviously in layer 5b) (Fig. 5 C and H; mutants: 54% ± 1.79%, P < 0.002), along with increased PLAP labeling in the thalamus (Fig. 6 J and O), suggesting that Ctip2 represses Tbr1. Because Ctip2 is genetically downstream of Fezf2 and because the increase in Tbr1 expression is similar in Fezf2 and Ctip2 mutants, it is possible that Fezf2 regulates Tbr1 expression indirectly via Ctip2.

Fig. 5.

Satb2 and Ctip2 regulate Tbr1 expression in a layer specific manner. (A–D) Immunostaining for Tbr1 (green) and Bhlhb5 (red) in P4 control cortex (A) reveals heavy staining in layer 6 and moderate staining in layer 5. A small fraction of cells in the upper layers are also Tbr1⁺. (B) Satb2 conditional knockouts show reduced Tbr1 in layers 6 and 5, and a loss of Tbr1 in the upper layers. Bhlhb5 expression is increased in the upper layers. (C) In Ctip2 mutants, Tbr1 staining increases in layers 6 and 5. (D) Ctip2^−/−;Satb2^LacZ/LacZ double knockouts at P0 show reduced Tbr1 expression in layer 5 and increased Bhlhb5 in the upper layers. (E and F) Immunostaining for Tbr1 (green) and Ctip2 (red) reveals an increase in Tbr1 and a decrease in Ctip2 expression in layer 5 of Fezf2^PLAP/PLAP mutants. Ctip2 (red) but not Tbr1 (green) expression is restored to control levels in layer 5 of Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double knockouts (F). (A–F) Each panel is a composite of tiled confocal images to form the full figure. (Scale bar, 100 μm.) (G–I) Graphs depicting the percentage of Tbr1+ neurons in different layers of the genotypes under study. Error bars are average ± SEM (n = 3 different animals). Confidence levels were calculated using a t test (*P < 0.5; **P < 0.05; ***P < 0.005, relative to controls).

Fig. 6.

Innervation of thalamus by PLAP⁺ axons correlates with Tbr1 expression. (A, B, and K) PLAP⁺ control brains revealed robust PLAP staining in the striatum (st) and thalamus (Th). (C, D, and L) Satb2 mutants show a reduction in PLAP⁺ axons traversing the striatum but almost no expression in the thalamus. (E, F, and M) Fezf2 mutants show robust PLAP staining in the striatum, anterior commissure (AC), and thalamus. (G, H, and N) In Fezf2^PLAP/PLAP;Satb2^LacZ/flox;Emx1-Cre double mutants, PLAP staining is reduced in the striatum and thalamus but remains robust in the anterior commissure. (I, J, and O) Robust PLAP staining is observed in the striatum and thalamus of Ctip2^−/− mutants. (A–J) PLAP innervation using colorimetric substrate for alkaline phosphatase. (K–O) Immunostaining using an anti-PLAP antibody to visualize PLAP⁺ tracts. (Scale bar, 200 μm.) All images are composites of tiled confocal images to show the entire section. (P) Genomic map representation of three MAR (matrix attachment regions) sequences identified in the Tbr1 genomic locus. (Q) ChIP analysis of chromatin from P0 WT neonatal cortex immunoprecipitated with an antibody to Satb2 using multiple primer sets targeting the Tbr1 MAR regions. These experiments show comparable levels of enrichment relative to Ctip2 A5 MAR [Ctip2 42/44 primer set (3)] region. Primers were normalized to a gene desert region. Each bar represents enrichment of the region using a distinct primer set designed to that region. Each primer set was tested at least three times in three independent samples.

The genetic relationship between Satb2 and Ctip2 led us to predict that Tbr1 expression should decrease in Satb2 mutants, owing to the up-regulation of Ctip2. The number of Tbr1⁺ neurons in layer 5 of Satb2 mutants is significantly decreased (Fig. 5 B and H; 3% ± 1.2%, P < 0.004) relative to controls, and PLAP innervation of the thalamus is reduced in Satb2 mutants (Fig. 6 D and L). Interestingly, in Fezf2^PLAP/PLAP;Satb2^LacZ/LacZ double knockouts, we observe PLAP⁺ axons in the thalamus, although the extent of innervation qualitatively seems to be less than in Fezf2 mutants but more than in Sabt2 mutants (Fig. 6 G, H, and N). Indeed, Tbr1 expression in layer 5 neurons of double knockouts is significantly increased compared with Satb2 mutants (Satb2^LacZ/LacZ: 3% ± 1.2%; Fezf2^PLAP/PLAP;Satb2^LacZ/LacZ double mutants: 12% ± 1.2%; P < 0.02). Satb2 mutants also show a nearly complete loss of Tbr1⁺ neurons in the upper layers (Fig. 5G; control: 23% ± 7.2%, mutant: 4% ± 0.3%; P < 0.01). There is a small but not significant change in layer 6, where Satb2 expression is relatively low (Fig. 5I; control: 63% ± 7.8%, mutant: 49% ± 5.4%; P < 0.1). Taken together, these data provide evidence that Tbr1 is regulated negatively by Ctip2 (and Fezf2).

Counting the number of Tbr1⁺ neurons in layer 5 of Ctip2^−/−;Satb2^LacZ/flox;Emx1-Cre double knockouts yielded a surprising result. Although there are more Tbr1⁺ cells in layer 5 of the double knockouts compared with Satb2 single mutants (Fig. 5H; 8% ± 1.8% in double mutants and 3% ± 1.2% in Satb2 mutants, P < 0.005), the numbers are neither restored to WT levels (22% ± 4.8%) nor to that of Ctip2 single mutants (54% ± 1.8%), suggesting that Ctip2 is not the sole regulator of Tbr1 expression and that Satb2 might positively regulate Tbr1. Moreover, Ctip2^−/−;Satb2^LacZ/flox;Emx1-Cre double knockouts fail to show a restoration of Tbr1 expression in the upper layers (Fig. 5 D and G), further providing evidence that Satb2 is required for the expression of Tbr1 in these neurons (independent of Ctip2). These studies suggest the possibility that Satb2 might bind directly to the Tbr1 genomic locus and promote its expression.

Satb2 Binds Directly to the Tbr1 Genomic Locus.

To determine whether Satb2 can bind directly to the Tbr1 locus, we identified three so-called MAR sites (Matrix Attachment Sequences) in the Tbr1 locus that have the potential to serve as Satb2 binding sites (Fig. 6P). To ascertain whether Satb2 can bind to any of these sites, we performed ChIP using an antibody against Satb2 (raised in the laboratory of Y.K. and T.K-S.) and chromatin from the cortices of WT mice at P0. As a positive control, we tested previously published sequences in the Ctip2 genomic locus for binding (Fig. 6Q) (9). As a negative control, we tested genomic sequences in a known gene desert region, which should show minimal or no interaction with Satb2. As expected, the negative control gene desert region showed minimal or no enrichment for Satb2 binding, whereas MAR sequences from the Ctip2 locus showed greater than sevenfold enrichment (when normalized to values observed in the gene desert region) (Fig. 6Q). We then used three primers specific to each MAR region in the Tbr1 genomic locus and asked whether these showed enrichment for Satb2 binding. All three sequences resulted in binding levels similar to that for the published Ctip2 sites. These data suggest that Satb2 can directly bind to the Tbr1 locus, thereby positively regulating the expression of Tbr1.

Regulation of Callosal Projections in Satb2 Mutant Neurons.

Until this point we have focused on the genetic interactions that regulate the formation of subcerebral and subcortical connections by deep layer neurons. Here we consider the interactions that mediate corticocortical connectivity as exemplified by the formation of axons that cross the CC, which we visualized using the Satb2^LacZ allele (9).

Satb2 mutants exhibit a failure of callosal development and the extension of β-gal⁺ axons to subcerebral targets, accompanied by a marked up-regulation of Ctip2 staining throughout the cortex (9, 10), whereas Fezf2 expression is unaltered. The ectopic expression of either Ctip2 or Fezf2 in WT upper layer neurons can divert their projections toward subcortical targets (8, 9), suggesting that the redirection of β-gal⁺ axons from callosal to subcerebral targets in Satb2 mutants may be due to the up-regulation of Ctip2. We therefore set out to explore the genetic interactions between Satb2 and Ctip2 in the formation of cortical connectivity. If the upregulation of Ctip2 is responsible for the alteration in axonal projections, then the loss of Ctip2 in a Satb2 mutant background should result in one of two possible outcomes: (i) the callosal projections of Satb2 mutant neurons will be rescued and β-gal⁺ axons will again cross the midline, or (ii) a novel default fate (neither callosal nor subcortical) might be revealed.

In Fezf2 and Ctip2 single mutants [Fezf2^PLAP/PLAP;Satb2^LacZ/+ (Fig. S2F) or Ctip2^−/−;Satb2^LacZ/flox (Fig. 1H and Fig. S2H)], β-gal⁺ axons from Satb2-expressing neurons cross the CC. As expected, in Fezf2^PLAP/PLAP;Satb2^LacZ/LacZ double knockouts, β-gal⁺ axons are absent from the CC and instead extend subcortically to the thalamus (Fig. S2K) and cerebral peduncle (Fig. S2L), similar to Satb2 single mutants (Fig. S2 C and D). These data confirm that Fezf2 is not responsible for the fate change in Satb2 mutant neurons.

LacZ staining of P0 brains from Satb2 conditional mutants reveals that β-gal⁺ axons enter the pyramidal tract (Fig. 1 B, F, and G), as expected. In contrast, β-gal⁺ axons are absent from the pyramidal tract of Ctip2^−/−;Satb2^LacZ/flox;Emx1-Cre double mutants (Fig. 1C), similar to control brains (Fig. 1A). Further, β-gal⁺ axons cross the CC (Fig. 1I) in these double mutants. These results suggest that the acquisition of Ctip2 expression by neurons that lack Satb2 is responsible, at least in part, for their failure to form callosal connections and the extension of β-gal⁺ axons to subcerebral targets (Fig. S1D). Interestingly, in Ctip2;Satb2 double mutants, many β-gal⁺ neurons still continue to project subcortically to the striatum, cerebral peduncle, or thalamus (Fig. 1 J, K, and L), although none reach the pyramidal tract (Fig. 1 L and M). Thus it seems that, without both Satb2 and Ctip2, neurons that normally express Satb2 may project callosally (owing to loss of Ctip2) or subcortically to the peduncle.

Overexpression of Tbr1 Can Rescue Callosal Projections in Satb2 Mutants.

Tbr1 plays an important role in specifying the corticothalamic projections of layer 6 neurons (5, 7, 17), but upper layer neurons that normally express Tbr1 do not project to the thalamus. This suggests that Tbr1 might have an unexplored role in the upper layers. Indeed, in Tbr1^−/− mutant cortices, the axons of upper layer neurons fail to cross the CC and instead terminate in Probst bundles at the midline (17). Interestingly, restoring expression of Tbr1 in the upper layers of Tbr1^−/− mutants did not direct those axons to the thalamus (7); instead they projected across the CC, consistent with the possibility that Tbr1 in upper layer neurons plays a role in callosal connectivity. We therefore asked whether we could rescue the callosal defects in Satb2^LacZ/LacZ mutants by reintroducing Tbr1 into the upper layers. Co-electroporation of expression constructs encoding Tbr1 and GFP into E15.5 Satb2^LacZ/LacZ mutants does rescue the formation of callosal projections: β-gal⁺ axons cross the CC in all 10 electroporated mutant embryos (Fig. 7 C and D), suggesting that expression of Tbr1 can compensate for the loss of Satb2. This defines a new function for Tbr1 and places Tbr1 genetically downstream of Satb2 in callosal neuron specification. Thus, in the developing cortex, Tbr1 has a dual function, first specifying layer 6 corticothalamic neurons and later in callosal connectivity.

Fig. 7.

Tbr1, EphA4, and Unc5h3 are downstream targets of Satb2 and can direct callosal projections. β-Gal+ axons fail to cross the CC in Satb2^LacZ/LacZ mutants electroporated with either a GFP control construct (A and B) or a PlxnA4 expression construct together with GFP (I and J). β-Gal⁺ axons cross the CC in Satb2 mutants electroporated with TBR1-IRES-GFP (C and D), EphA4 and GFP (E and F), or Unc5H3 and GFP (G and H). E and F are composites of tiled images to form the full figure. (Scale bar, 100 μm.) (K–N) Auts2 expression is lost in the upper layers of Satb2 mutants. (K and L) Auts2 protein (green in K, white in L) is expressed in both the upper and deep layers of control Satb2^LacZ/+ brains at P0. Ctip2 expression is shown in red. (M and N) Auts2 expression is maintained in layer 6 of Satb2 mutants but is down-regulated in the upper layers. (Scale bar, 50 μm.) (O) Model of genetic interactions between Fezf2, Ctip2, Satb2, and Tbr1 in wild type cortex. Fezf2 expression in layer 5 neurons represses Tbr1 and Satb2, which repress corticothalamic and callosal fates, respectively. The repression of Satb2 enables expression of Ctip2 and Bhlhb5, which are required for the specification and execution of a subcerebral identity. In Satb2 expressing neurons, Ctip2 is repressed, leading to a repression of subcerebral identity and acquisition of callosal and corticocortical axonal projections.

Loss of Tbr1 Expression in Satb2 Mutants Coincides with a Loss of Auts2.

Tbr1 binds to and directly activates the expression of Auts2, a gene expressed in frontal cortex that has been linked to autism and mental retardation (18). The functional significance of this relationship has been puzzling because Tbr1 regulates corticothalamic identity, and there have been no reports of defects in corticothalamic tracts in autistic patients. However, imaging studies in patients with autism have revealed defects in callosal tracts (2, 19–26). A recent publication highlights a deficit in long-range connections (such as callosal connections) and excess of short-range cortical connections in patients with autism spectrum disorder (27).

Auts2, like Tbr1, is expressed in both the deep and upper layers of the cortex (Fig. 7 K and L). Because Tbr1 regulates Auts2, we investigated whether the loss of Tbr1 expression in upper layer neurons in Satb2^LacZ/LacZ mutants coincides with changes in the expression of Auts2. We observed a striking loss of Auts2 expression in the upper layers of Satb2 mutants (Fig. 7 M and N), similar to the loss of Tbr1 in Satb2 mutants (Fig. 5B); there was no change in Auts2 expression in layers 5 or 6 (Fig. 7 M and N) relative to controls (Fig. 7 K and L). These data are consistent with the possibility that Satb2 regulates the expression of Tbr1, which in turn is required for Auts2 expression in callosal projection neurons. These results may have implications for the etiology of autism.

Expression of EphA4 and Unc5H3 Restores Callosal Projections in Satb2 Mutants.

Previously we identified several genes that show altered expression in Satb2^LacZ/LacZ mutants (9). In particular, three axonal guidance molecules (EphA4, PlxnA4, and Unc5H3) are down-regulated in upper layers of mice lacking Satb2. Prior studies have implicated Ephs and ephrins in callosal development (13, 28, 29). EphA4 is normally expressed in upper layer callosal neurons and the glial wedge (28). In Satb2 mutants, EphA4 expression is lost in cortical neurons, but expression in the glial wedge is maintained (9). Unc5H3 mutants have no reported callosal deficiencies (13), but mice lacking Netrin, a ligand for Unc5H3, lack both the CC and the anterior commissure (29, 30). To test the hypothesis that one or more of these genes is required for the proper guidance of callosal axons to their destinations, we attempted to rescue the formation of callosal projections in Satb2 mutants by reintroducing the expression of individual axon guidance molecules into upper layer neurons. In utero electroporation of PlxnA4 failed to rescue callosal projections (Fig. 7 I and J) in Satb2^LacZ/LacZ mutants, but electroporation of EphA4 (Fig. 7 E and F) or Unc5H3 (Fig. 7 G and H) resulted in the extension of β-gal⁺ axons across the CC. These results suggest that EphA4 and Unc5H3 are critical downstream targets of Satb2 in callosal fate specification.

Discussion

Our data suggest that cortical projection neurons actively repress alternate fates to promote appropriate fate choices during development (Fig. 7O). When specific repressive interactions are removed, alternative fates are executed (Fig. S1). Whereas Tbr1, Ctip2, and Satb2 are expressed in postmitotic neurons, Fezf2 is expressed in cycling cortical progenitors from very early stages of corticogenesis (11). Loss of Fezf2 is critical for the specification of the subcerebral projections of layer 5 neurons, as evidenced by the loss of corticospinal projections in Fezf2 mutants (11, 14). During the earliest stages of corticogenesis, Sox5 expression in subplate and layer 6 neurons represses the expression of Fezf2 (and consequently that of Ctip2) (15, 16). This likely promotes the expression of Tbr1 in layer 6 neurons. Tbr1 binds to and represses the Fezf2 genomic locus (5, 7), thereby suppressing a subcerebral fate and promoting the formation of corticothalamic projections from layer 6. Sox5 expression is down-regulated early in layer 5 neurons (15, 16), leading to a derepression of Fezf2, which consequently leads to a repression of Satb2. The effect of this is threefold. First, there are no intracellular triggers to promote a callosal fate. Second, the absence of Satb2-mediated repression of Ctip2 and Bhlhb5 leads to the continued expression of these genes and the extension of subcortical axons. (Interestingly, although Bhlhb5 expression in Satb2 mutants at E18 does not differ from that in controls, expression at P4 is significantly increased in Satb2 mutants. This suggests that Satb2 is required for the normal down-regulation of Bhlhb5 expression during early prenatal development, and that the sustained expression of Bhlhb5 in Satb2 mutants might be important in executing or maintaining the new subcerebral fate of these neurons.) Third, our data indicate that the absence of Satb2-mediated activation of Tbr1 suppresses corticothalamic projections. Thus, Fezf2-expressing layer 5 neurons extend their axons subcerebrally.

We hypothesize that during production of the upper layers, the absence of Fezf2 in cortical progenitors allows their daughters to express Satb2, which in turn promotes a callosal identity (in part, surprisingly, by activating Tbr1 in upper layer neurons). Simultaneously, activation of Satb2 results in the repression of Ctip2 and Bhlhb5, ensuring that executors of subcortical identity remain inactive in callosal neurons. Thus, each phase of corticogenesis and neuronal fate specification deploys an active repression of previous fates and a promotion of the appropriate layer-specific projection fate (see SI Discussion for further details).

Tbr1 seems to play distinct roles at different stages of cortical development. At early stages, Tbr1 promotes a frontal identity while suppressing caudal identity (4). During the formation of layer 6, Tbr1 plays an essential role in specifying the fates and projection patterns of corticothalamic neurons (4, 5, 7, 17). Interestingly, although corticothalamic projections are decreased in Tbr1 mutants and increased in Fezf2 mutants (which show an up-regulation of Tbr1), these alterations in projections are incomplete: there are still some corticothalamic axons in Tbr1 mutants, and Fezf2 mutant neurons do not completely convert to corticothalamic identity, suggesting that Tbr1 cannot be the sole specifier of a corticothalamic fate.

In addition, we found that Satb2 and Ctip2 dynamically regulate the expression of Tbr1 in a layer dependent manner. Satb2 seems to promote expression of Tbr1, whereas Ctip2 represses Tbr1 expression. This complex bidirectional regulation might explain why we observe a partial rescue of callosal axon targeting in Ctip2;Satb2 double mutants (Fig. S1F). In these animals, Tbr1 expression in layer 5 neurons is partially rescued (relative to Satb2 single mutants) and might account for the partial rescue of callosal projections in these animals. In the upper layers, Tbr1 seems to play a unique role in promoting the formation of callosal projections, which may explain why Tbr1 mutants display callosal agenesis (17). This role for Tbr1 in callosal development is also intriguing in the context of autism. Autistic patients seem to have callosal abnormalities (19–26), and in this context it is relevant that Auts2, a gene associated with autism, is regulated by Tbr1, which in turn is regulated by Satb2, a callosal fate specification gene. Recent studies suggest that a mutation in CAv1/2 (an L-type calcium channel) that is associated with Timothy syndrome, a form of autism, results in lower numbers of Satb2-expressing cells and an increase in Ctip2-expressing cells (31). The authors suggest that the reduction in callosal neurons in Timothy syndrome is consistent with the emerging view that autism spectrum disorders arise from defects in connectivity between cortical areas (31).

The role of axon guidance molecules, specifically Ephrins and their receptors, in directing callosal neurons has been studied extensively using genetic knockouts. Although the loss of single genes does not result in dramatic callosal abnormalities, double knockouts of various genotypes do compromise callosal development, suggesting that there is redundancy built into the system. Indeed, it is interesting to note that although restroring EphA4 expression in Satb2 mutants rescued callosal projections, EphA4 knockout mice fail to display callosal defects (28). However, multiple EphB receptors (B1, B2, and B3) and ligands (ephrinB1, B2, and B3) are expressed in callosal fibers and midline guidepost cells, respectively, and it is likely that their functions are redundant. Restoring expression of Unc5h3 also rescued callosal projections in Satb2 mutants. Unc5H3 is normally expressed throughout the cortical plate during development, and no callosal defects have been observed in Unc5h3^rcm null mice, although these mice display subcortical defects in CST development (13). Our findings suggest an unexpected role for this receptor in callosal development.

Methods

Animals.

Satb2^LacZ/+, Fezf2^PLAP/+, and Ctip2^+/- animals were mated to generate appropriate double knockouts. Details of generation of the conditional Satb2 single and double mutants are described in SI Methods.

Immunohistochemistry.

Standard immunohistochemistry techniques were used. Details of antibodies used in the study are listed in Table S1. PLAP staining was performed as described previously (11).

In Utero Electroporation:

Methods are as described in previous work (8).

ChIP.

Experiments were performed as described earlier (9) using a rabbit anti-Satb2 antibody. MAR sequence predictions were based on www.genomatix.de. For primer sequences, refer to Table S2.